节点文献
基于车载激光扫描数据的地物分类和快速建模技术研究
Research on Methods for Ground Objective Classification and Rapid Model Construction Based on Mobile Laser Scanning Data
【作者】 喻亮;
【导师】 詹庆明;
【作者基本信息】 武汉大学 , 地图学与地理信息系统, 2011, 博士
【摘要】 城市三维建模是“数字城市”的重要组成部分,其不仅为城市管理者提供直观的虚拟城市模型以方便城市规划、城市建设、公共安全、公益服务等各项工作,同时也为普通民众提供更加方便数字化生活的载体。车载激光雷达测量系统是近年来发展起来的城市三维建模的新技术,为建立更加逼真的城市三维模型提供了新途径。本文旨在研究基于车载激光扫描数据进行城市建模的过程、难点和方法,主要进行了以下几方面的工作:1.回顾了车载激光雷达测量系统的发展及其数据特点、数据处理技术的进展概况。车载激光扫描测量系统是在传统的传统激光雷达系统基础之上发展起来的以汽车作为平台实现多传感器集成的测量系统。与机载激光扫描系统相比,车载系统能够获取的建筑物立面数据精度更高;与单站激光扫描系统相比,车载系统能够获取车行路线上多个测点的拼接扫描数据和影像信息,是当前迅速发展城市街景数据获取手段。2.融合点云不同特征进行点云分割的方法研究:激光扫描点云数据包含扫描对象表面的三维坐标、回波强度等信息,通过和CCD相机获取的影像数据的配准解算,还可以获取扫描点的颜色信息。本文在研究当前点云分割的基于几何特征和基于光谱特征的方法基础之上,将点云的几何特征和影像特征综合考虑为点云的特征向量空间的维度特征,提出了基于多维欧几里得向量空间的临近度判别方法,并对维度特征引入权值系数来反映不同特征对点云划分的影响程度,提高了该算法对不同来源的点云数据的可靠分割质量和效率。3.复杂城市场景中的地物分类识别方法研究:车载激光扫描系统获取的城市街道环境的点云数据中既包含规则地物(地面、建筑物里面、交通标识牌),也包含不规则地物(花坛、行道树、行人等),本次研究首先从车载系统获取的场景数据中包含的地物对象人工模式化入手,总结不同地物在构造形态和空间分布上的规则;然后以点云分割结果为基础,将点云面片作为分类识别的最小处理单元,采用面向对象的设计方法建立点云面片对象类,并总结出一套面片对象的属性和方法体系,从而在地物特征和点云面片对象属性之间建立关联关系,考虑到以往的基于知识的点云分类方法的刚性判断模式的不足,提出基于特征模糊度量和可信度判断的柔性分类模式,提高了对复杂场景中的地物分类识别的精度。4.基于OpenGL的城市地物快速三维建模的方法研究:首先分析了三维建模中边界模型(Boundary Representation)和实体模型(CSG)的差异,然后简要介绍了OpenGL建模的基本方法,并实现了基于数据驱动的对象三维重建方法。针对城市街景中地物的特征,重点介绍了基于凸包生成的方法构建规则建筑物平面的方法,以及利用Hough变换和最小二乘法提取穹顶、圆柱等弧状表面的对象特征并建模的方法,并对建成的三维模型进行纹理贴图表现,以提高模型的可视化效果。5.开发出点云数据处理原型系统:利用本次研究所取得的主要成果,开发出点云数据处理原型系统对上述算法进行试验处理。并利用当前已获得的多套机载激光点云数据、单站激光点云数据、车载激光点云数据分别对点云分割算法、点云分类识别算法和三维模型重建方法进行了对比分析。
【Abstract】 Urban3D model is an important part of "Digital City" that not only offers a virtual simulation for urban planners in urban planning, facilities construction, public security and commonweal services, but also serves as a convenient platform for public participation. Mobile Laser Scanning System is an innovative urban3D modeling technology developed recently that is often used to construct more realistic urban3D models. This dissertation focuses on issues related to processes, difficulties and methods of urban model construction based on Mobile Laser Scanning data. The contents of this research consist of the following parts:Literature reviews and summaries of development of Mobile Laser Scanning System, data features and data processing technology. Mobile Laser Scanning System develops based on traditional radar system and becomes a new measuring system that installed on vehicle platform and integrated with multiple sensors. Compared with Airborne Radar Scanning system, the Mobile Laser Scanning System can obtain higher resolution data of building facade. Moreover, compared with Single-Spot Laser Scanning system, the Mobile Laser Scanning System is capable of scanning combined image and information from different scanning spots by navigating along urban roads, which becomes main methodology of current data collection of street view.Point cloud segmentation integrated with different data features:Laser Scanning point cloud data includes3D coordination of object surface, echo intensity and etc. The color information of scanned object can also be obtained by calibration integrated with data that captured by CCD camera. Based on current methodology of point cloud segmentation from topology and spectrum features, this dissertation has addressed methodology based on proximity judgment on Euclidean distance in multiple dimensions while considered both topology and image features as dimension features of point cloud comprehensively. Furthermore, a weigh coefficient has also been introduced to reflect the influence degree of segmentation from different features. Therefore, the creditability of this algorithm in calculating data from different resources has been enhanced.Object classification in complex urban scene:the Vehicle-Born Radar Scanning system can capture the point cloud data of urban street with regular forms (urban ground, inside of building and traffic sign) and irregular forms (parterre, trees and passengers). At first, this research artificially modularizes the data that is captured by Vehicle-Born Radar Scanning system and also includes ground objects. The regulations of different objects in forming and spatial distributing are also summarized. Secondly, based on the segmentation results, point cloud surface has been set as the minimal recognizing unit in classification; the Objective Oriented Design has been utilized to establish point cloud surface class; the attribution and method of such objective has also been introduced. Therefore, the relationship between urban ground objective and point cloud surface attribution has been established. To avoid the disadvantage of previous arbitrary modularization in point cloud classification, a fixable modularization based on fuzzy feature measurement and credibility judgment has also been addressed. The precision of ground object classification in complex urban scene has been improved accordingly.3D urban model quick construction based on OpenGL:the difference between Boundary Representation model and Constructive Solid Geometry model in3D model construction has been analyzed initially. The basic methodology of OpenGL model construction has also been introduced, and the3D model reconstruction has been implemented based on Data Driven methodology as well. In the following phrase, this research has concentrated on Convex Hall Generation method to construct regular building plane and Hough Transformation and Least Squares method to construct the arc features including dome and pillar, concerning the unique characteristics of urban objectives. The texture has been attached to enhance the realistic simulation.The prototype of point cloud data processing system:The achievements of this research have implemented into a prototype of point cloud data processing system to testify those algorithms. The comparison between point cloud segmentation algorithm, point cloud classification algorithm and3D model reconstruction has been employed by utilizing the existing data that includes Airborne Laser Scanning data, Terrestrial Laser Scanning data and Mobile Laser Scanning data.
【Key words】 LiDAR; laser scanning point cloud; point cloud segmentation; pointcloud classification; 3D model reconstruction;